Stationary Phase Integrals, Quantum Toda Lattices, Flag Manifolds and the Mirror Conjecture

Stationary Phase Integrals, Quantum Toda

Lattices,

Flag Manifolds and the Mirror Conjecture

Alexander Givental *
UC Berkeley

July 8, 1996, Revised December 1, 1996

0. Introduction. Consider the differential operator

h 2 A d 2

=o

This operator is a quantization of the Hamiltonian of the Toda lattice on
n + 1 identical particles with configuration coordinates to, ...,t n and with the
exponential interaction potential exp(tj — of neighbors. The Toda lattice
is known to be integrable on both classical and quantum levels: there exist
commuting differential polynomials D m (hd/dt,expt,h),m = 0,...,n, which
play the role of quantum conservation laws (i.e. [H, D ] = ... = [H, D n ] = 0)
and whose symbols D m (p, exp t, 0) form a complete set of Poisson-commuting
first integrals of the classical Toda lattice. In this paper we study solutions
S(t) of the differential system DqS = ... = D n S = whose characteristic
Lagrangian variety L is the most degenerate invariant Lagrangian variety

of the Toda lattice. According to [12|, |13| this Lagrangian variety is the

spectrum of the quantum cohomology algebra of the manifold of complete
flags C C 1 C ... C C" +1 . We represent solutions S by stationary phase
integrals in n(n+l)/2 complex variables and point out the role these solutions
play in the quantum cohomology theory. As we explain in the last section,
our results prove the mirror conjecture in the case of the flag manifolds.

*Research supported by NSF grant DMS-93-21915 and by Alfred P. Sloan Fellowship.

1

1. The Toda D-module.

Denote Do, D n coefficients of the polynomial

\ n+1 + D X n + ... + D n =

det

X+Po qi

-1 X + Pi q 2
-1 X+p 3 q 3

-1 X + p n

Do = Y;Ph D i= Y.i>jPiPj + Y. ( liT-- ■ The polynomials D m {p,q) oip ,...,p n
andQgi = exp(t! — i )> q n = exp(t n — t n _i) form a complete set of Poisson-
commuting first integrals of the Toda lattice (see for instance |12| ). Their
quantizations

D m (hd/dt , . . . , hd/dt n , exp (t i - t ) , . . . , exp(t n - t n _i ) ) ,

m

0.

n.

are defined unambiguously since any monomial in D m containing q^ contains
neither pi nor Pi-\.

Theorem 1. [H,D ] = ...= [H, D n ] = 0.

Proof. The commutator [H, A] of the Hamiltonian operator H with the
above determinant A = X n+1 + DoX n + ... vanishes modulo h 2 since symbols of
D m Poisson-commute with the symbol — ^ q^ of H. Also [H, A] does

not contain any terms of order higher than h 2 since for any 3 distinct i, j, k
we have d 3 qi/dtidtjdt k = 0. Computing symbols of the terms proportional
to h 2 we find that the contribution of Agj is equal to —qid 2 A(p, q) / dpidpi-i,
and the contribution of h 2 d 2 A/2dt 2 is qid 2 A/2dpidpi^i+q i+ id 2 A/2dpidpi + i.
After summation over i all these contributions cancel out. □

We will study solutions of the PDE system

DnS

D n S = (and hence HS = 0),

i.e. the solution sheaf of the left module V/V(D , D n ) over the algebra
T> of differential operators D(hd/dt,expt, h) with Fourier-polynomial coeffi-
cients. We introduce the symbol D(p,expt,0) of the differential polynomial

1 Throughout Sections 1 - 3 we will systematically use the notation \ogqt for ti — U-i.

2

D and call the Lagrangian variety L C T*(C — 0) n+1 given by the equations
Do(p,q) = ■■■ = D n (p,q) = the characteristic Lagrangian variety of this
P-module. According to [|T7| the Lagrangian variety L is nonsingular. The
operator D m is weighted-homogeneous of degree m + 1 with respect to the
grading degft = 1, degg^ = 2, degtj = and therefore L is also weighted-
homogeneous with weights degp^ = 1, degg^ = 2. For generic t the fibers
L fl T * xpt of the projection L — > (C — 0) n+1 consist of (n+ 1)! distinct sim-
ple points. This follows from Sard's lemma, and also can be deduced by
induction from the continued fraction formula for the determinant A which
coincides with the numerator of the following rational function of A:

A + Po H — ■

A + ^+ | q n -

A +Pn

2. The stationary phase integrals. Consider the following "2 - di-
mensional Toda lattice" with (n + l)(n + 2)/2 vertices and n(n + 1) edges:

• — > •

• — > • — > •

For each edge a of the lattice we introduce a factor Q a and introduce an-
other, excessive notation Ui (respectively Vi) for the factors Q a corresponding
to the vertical (respectively horizontal) edges next to the diagonal boundary
of the lattice as shown on the diagram. We denote Y the affine algebraic va-
riety in the n(n+ l)-dimensional complex space with coordinates Q a given by
the n(n — 1)/2 equations Q a Qp = Q^Qs making the diagram "commutative":

3

one equation for each lxl cell

Qa

• — > 9

<2 7 I I Qf3 ■

• — > •

Qs

We put q 1 = u±Vi, ...,q n = u n v n and denote Y q the n{n + l)-dimensional fibers
of the map from Y to C n defined by these formulas.

If qi 7^ for alH = l,...,n, the relations between Q a allow to express
them via the set {T u , v = < j < i < n} of vertex variables:

Qa = exp(T ;y+ ( Q ,) - T v _(a)),

where i/-(a) and v + (a) are respectively the indices of the source and target
vertices of the edge a. In particular, q^ = exp(tj — ti-\) where to — Too,t\ =
Tu, and the fiber Y q is isomorphic to the n(n+ l)/2-dimensional complex
torus with coordinates {expT^, < j < i < n}.

On Y q with q G (C — 0) n we introduce the holomorphic volume form

u q = K=i A}^ dT l3 ,

the holomorphic function T q obtained by restriction to Y q of "the total Toda
potential energy"

edges a

and the stationary phase integral

Sr(t) = [ e^ h oo q .

JTcY q

In this definition, Y represents a (possibly non-compact) cycle in Y q of middle
dimension such that the integral converges. For h > and generic q one can
construct such a cycle by picking a non-degenerate critical point of T q and
taking the union of descending gradient trajectories of the function Re^-g
with respect to a suitable Riemannian metric on Y q on the role of T. The
stationary phase integral depends only on the homology class of the cycle in

4

the appropriate homology group Ti. q which can be described as the inverse
limit as M — ■> oo of the relative homology groups

H dimcY «{Y q , {Ref q < —M}).

The rank of the group H q is equal to the number of critical points of T q
for generic q since all critical points of the real part of a holomorphic Morse
function have the same Morse index. The Gauss-Manin parallel transport
of cycles identifies the groups 7i q for close q but may give rise to a nontriv-
ial global monodromy. The notation Sr{t) emphasizes the multiple- valued
character of the stationary phase integrals which therefore depend on the co-
ordinates log qi — ti — ti-i on the universal covering of the parameter space.
The dependence of the integral on h is suppressed in this notation.

Theorem 2. The stationary phase integrals Sr(t) satisfy the differential
equations D S — D±S — ... — D n S = 0.

Proof. Application of the differential operator A = A n+1 + D X n + ... to
the stationary phase integral produces the amplitude factor e'^^Ae^^ —
det(A + A) where

A =

-Mi U\Vi

— 1 Vi — u 2 u 2 v 2
-1 v 2 - u 3 u 3 v 3

-1

We should show that this amplitude is congruent to A n+1 modulo linear
combinations of Lie derivatives along vector fields Q m d/dT u tangent to Y q ,
i.e. linear combinations of hdQ m / dTij + Q m dT jdT^ with j < i. Notice that
0F/dT u is the sum of Q a over the (< 2) edges a ingoing the vertex v minus
such a sum over the outgoing edges.

We begin induction on n by noticing that for n = 1 the amplitude is
equal to A 2 — XdJ- ' j&T 2 \ and apply the induction hypothesis to the triangular
lattice with the principle diagonal i — j cut off.

Consider the differential operator A' defined as the determinant of the
3-diagonal nxn-matrix with X+hd/dT^i, i — 1, n, on the principal diag-
onal, — l's under the diagonal and Vj-Uj+i, % = 1, n — 1, above the diagonal.
Denote T' the sum of all Q a except u^s and i^'s . By the induction hypothe-
sis (and "commutativity" Q a Qp = Q^Qs of the squares next to the diagonal

5

% = j) we may assume that the amplitude factor exp(—J-'/H)A'exp(J-'/H) is
congruent to A n modulo Lie derivatives along Q m d/dTij with \i — j\ > 1.

The vector field Q m d/dT iti _i, with no edges adjacent to the vertex v =
(i, i — 1) present in the monomial Q m , produces the amplitude Q m dJ-'/dT it i_ 1 .
Since the vertices — 1) do not have common edges, addition of such am-
plitudes allows to transform the amplitude factor exp(— T'JK) A' exp(J r '/h)
purely algebraicly as if OF / dT^_i = 0. Using such transformations we can
replace dT' '/dT^-i by Vi — Ui, and the induction hypothesis can be refor-
mulated as the congruence to A n+1 of the (n + l)-determinant det(A + B)
where

B =

Vi — U\ V\U 2

-1 v 2 - u 2 v 2 u 3
-1 v 3 -u 3 v 3 u 4

-1

The matrices A and B admit the factorizations A = UV, B = VU into the
product of the following triangular matrices:

U

Ui

1

U 2
1

u 3

1

u. r .

V =

-1 V!

-1 v 2

^3

Thus A is similar to B and has the same characteristic polynomial. □

The family T q generates the following Lagrangian variety parametrized
by critical points of the functions (and responsible for stationary phase ap-
proximations to the integrals Sr(t)):

{(p, q) \3y E Y q : d y T q = 0, p = dF q (y)/dt}.

The identity det(A + A) = det(A + B) also proves by induction the following
Corollary 1. The Lagrangian variety generated by the family T q co-

incides with the invariant Lagrangian variety L
D n (p, q) = 0} of the Toda lattice.

{(p,q)\Dq(p,q)

6

Notice that the equations dTjdT v = of the critical points and the
notations pi = dF/dti can be interpreted, in the spirit of the elementary
theory of linear electric circuits, as homological boundary conditions for the 1-
chain ^ on the oriented graph. One can therefore describe the critical

points by the relations Q a = J$ + ( a ) ~ J<f>-(a) introducing the 2-chain J(t>[4>],
a linear combination of the n(n + l)/2 clockwise oriented 1 x 1-cells of the
lattice (the edge ±a occurs in the boundary of the cells ±<fi±(a), the index
= (i, j) runs 1 < j < i < n). Additionally, po = —Jn,pi = J\i — JriiPi =
J22 — J33, ■■■,Pn = Jnn- Now cancellations of all Jy with i > j in the total
sum ^2 Q a prove

Corollary 2. The generating function T q {y C rit) on the Lagrangian variety
L equals Yn=i = ~ n Po + ( 2 - n)pi + (4 - n)p 2 + ... + np n .

Remark. Corollary 2 can be also deduced (see ]12[) from the weighted
homogeneity of L.

3. Quantum cohomology of flag manifolds. The cohomology al-
gebra H*(F) of the flag manifold F = {0 C C 1 C C 2 C ... C C n+1 } is
multiplicatively generated by the 1-st Chern classes pi of the tautological
line bundles with fibers C* +1 /C\ A complete set of relations between the
generators can be written in the form (A + po)(A + pi)...(A + p n ) = \ n+1
equating the total Chern class of the sum of the tautological line bundles to
that of the trivial bundle with the fiber C n .

On the Poincare-dual language of intersection indices (•, •) the structural
constants (ab, c) of the cohomology algebra count (with signs) isolated com-
mon intersection points of the three cycles a, b, c in general position.

The quantum cohomology algebra of the flag manifold is defined as a
deformation of the algebra H*(F) with structural constants (a ob, c) counting
isolated holomorphic spheres (CP 1 , 0, 1, 00) — > (F, a, b, c) passing by the three
marked points through the three cycles. A precise definition can be based
on Gromov's compactness theorem and Kontsevich's concept |TJ| of stable
holomorphic maps.

Let S denote a compact connected complex curve with at most dou-
ble singular points. It is rational (i.e. if 1 (S,(9) = 0) if and only if all
irreducible components of S are spheres and the incidence graph of these
components is a connected tree. Denote x = (xi,...,Xk) an ordered set of
pairwise distinct non-singular marked points on E. Two holomorphic maps
(E,x) — > F, (E',0;') — > F are called equivalent if they are identified by a

7

holomorphic isomorphism (£, x) — > (£',£'). A holomorphic map — > F
is called stable if it does not admit non-trivial infinitesimal automorphisms.
For rational £ stability means that each irreducible component mapped to a
point in F carries at least 3 marked or singular points.

The degree d of the map £ — > F is defined as the total 2-nd homology

class of F it represents. According to [[14], ||, equivalence classes of stable
degree-<i holomorphic map of rational curves with k marked points to the flag
manifold F form a compact complex non-singular orbifold which we denote
Fj^d- If non-empty, it has the dimension dimF + (—Kp, d) + k — 3 where
— Kp is an anti-canonical divisor of F.

Evaluation of maps (£, x) — > F at Xj defines the evaluation maps :
Ffc,d ->• F, « = 1, k.

Examples. 1) The moduli space F 3>0 consists of classes of constant maps
(CP 1 , 0, 1, oo) — > F and thus is isomorphic to F. The moduli spaces F fe)0
with < 3 are not defined since constant rational maps with less than 3
marked points are unstable.

2) Forgetting the space C l in a flag defines the projection F — > Fj to the
partial flag manifold Fi with fibers isomorphic to CP 1 . Denote a« G H2(F)
the homology class of the fiber. In fact any compact holomorphic curve in
F of the degree a; is one of the fibers. This identifies the moduli space F 0>ai
with the base p.

3) The evaluation map e\ : Fi >a . — > P is an isomorphism.

4) The Borel-Weil representation theory for SP n +i(C) identifies the root
lattice of type A n with the Picard lattice H 2 (F) of the flag manifold. In
H 2 (F) the classes aj, a n form the basis of simple coroots dual to the basis
J\,...,J n G H 2 (F) of fundamental weights Jj = — po + ... — p^, i.e. Chern
classes of the tautological bundles A l (C 1 )*. The simplicial cone spanned in
H 2 (F, M) by the fundamental weights is the Kahler cone of the flag manifold.
This implies that the degree d of any compact holomorphic curve in P is a
sum d\ai + ... + d n a n with all dj > 0. The sum J2i>jPi~Pj = 2(Ji + ... + J n ) of
positive roots represents the anti-canonical class of the flag manifold. Thus
dimP 3jd = dimP + 2{d x + ... + d n ).

Let us introduce the grading in the algebra C[A] = C[qi, q n ] of the semi-
group A = {d = Y.di(ii G H 2 {F)\di > 0} by putting degg d = 4(d a + ... + d n ).
We extend C[A]-bilinearly the Poincare pairing (A, B) = j F A A B in the De
Rham cohomology of P to the graded C[A]-module H*(F, C[A]). Structural

8

constants of the C[A]-bilinear quantum multiplication o on H*(F,C[L]) are
defined by

VA, B, C E H*(F) (AoB,C) = S ^q d I e\{A) A e* 2 {B) A e*(C) .

deA

One can show (see e. g. [Q, 0]) that on H*(F, C[A]) the quantum multipli-
cation defines the structure of a commutative associative graded Frobenius
algebra with unity 1 e H*(F) (i.e. (A o 1, C) = (A, C) ). Modulo (q u g n )
this structure coincides with the cup-product on H*(F). Recent develop-
ments in symplectic topology and theory of stable maps allow to extend
the construction of the quantum cohomology algebra to arbitrary compact
symplectic manifolds.

The spectrum of a quantum cohomology algebra can be naturally identi-
fied with certain Lagrangian variety — the characteristic Lagrangian variety
of the quantum cohomology T> -module.

For k > denote c the 1-st Chern class of the tautological line bundle
over the moduli space Fk,d with the fiber at the point [(E, x) — > F] equal
to the cotangent line T* E to the curve at the first marked point. Denote
pt = Y17=oPrfi = XX^« — Ji+i)U the general 2-nd cohomology class of F and
dt = — U-i) the value of this cohomology class on the homology

class d = Y^didi. For each A e H*(F) the vector-function s^i) of t with
values in H*(F, C) is defined by

VB e H*(F) (s A (t), B) = (e pt/h A, B) + V e dt [ e 2^1^3 A e *( B )

,f^n J Fa,* K-C

deA-0

By the definition, sa is a formal power series of qi = exp(tj — with
vector-coefficients which are polynomial in log^j and frr 1 .

The vector series sa satisfy the following linear differential equations with
periodical coefficients (see e. g. [|, U, §):

* d

n—s A =PiOs A , % = 0, n,

where piO are operators of quantum multiplication by pi. In particular,
the equations are compatible for any value of h (i.e. pi o pj = pj o p i and
d(pi°)/dtj = d(pjo) / dti) , and the linear space of all solutions to this system

9

coincides with the space of all vector-functions sa (of dimension rk H*(F)
over, say, C((7i -1 )) ).

By definition, the quantum cohomology D-module is generated by the
scalar functions S^t) := (sA(t), 1), i.e. coincides with V/I where / = {D G
V\DS A = MA G H*(F)}. It is easy to show (see for instance [§) that if
a homogeneous differential polynomial D(fhd/dt,expt,h) annihilates all the
functions Sa then the relation D(p, q, 0) = holds in the quantum cohomol-
ogy algebra.

Examples. 5) By the very definition ^dS^/c^i = since J^pj = in
H*{F). ^

6) The relation p\ + ... + p\ = holds in the cohomology algebra H*(F)
of the flag manifold. We claim that in the quantum cohomology algebra
Pq 2 + ... + p° 2 = 2qi + ... + 2q n . Indeed, for the degree reasons Jj o Jj must
be equal to Jj Jj plus a linear combination A m g m . The coefficient X m is
equal to the number of holomorphic maps CP 1 — > F passing by 0, 1 and
oo trough the generic divisors Jj, Jj and a given generic point in F. From
Examples 2 and 4 we find therefore that Jj o Jj = J { Jj + 5 i3 -g f and thus

E K 2 /2 = E ^ 2 - E Ji ° J*-i = E a-

Theorem 3 O). TTie Hamiltonian operator H = {^h 2 d 2 / 'dtf) /2 —
exp(tj — tj_i) of the quantum Toda lattice annihilates the functions Sa, A G
7J*(F).

Proof. Application of the operator H to Sa yields

(H(po,q) s A A) + hJ2(^-SA,l) ■

i

The first term vanishes due to Example 6. Since (pi o B, 1) = (B,pi o 1) =
(B,pi) is constant for any £? G H*(F), we have ((dpiO /dU)sA, 1) = (sa, (<9pi°

/a*i)i) = o. □

Corollary. S A = (e pt/h s, A) where the coefficients s {d) G H*(F,C(h)) of
the vector-function s = EdeA s Q d can be found recursively from

n

s (0) = 1, H[H(d,d) + ^d i J l }s w = sid ~ ai) ■

i=l i:di>0

In particular, the formal series s converges everywhere.

10

Remark. Analogous computations for the flag manifold G/B of a semi-
simple complex Lie group G of rank n give rise to the Hamiltonian H =
^(aj, dj)JiJj/2 — XX a «; a «) ( ?i/2 of the Toda lattice corresponding to the sys-
tem of simple coroots ai,...,a n Langlands - dual to the root system of G.
The proof of the theorem below is a specialization to the case G = SL n+ i (C)
of the general results by B.Kim describing quantum cohomology P-modules
of flag manifolds G/B in terms of quantized Toda lattices.

Theorem 4. D S A = D^a = ... = D n S A = for all A G H*(F).

Proof (see ||13| ). Since HD m SA = D m HSA = 0, the formal power se-
ries S = Q d '■= D m SA with coefficients polynomial in logg (and fr l )
satisfies the hypotheses of the following

Kim's Lemma. If S<® = and HS = then S = 0.

Proof (see [O] ) uses only ellipticity of the operator H and polynomiality
of its coefficients. Consider a non-zero term S^q d of minimal degree and
pick in it a non-zero monomial term const - (log q) m q d of maximal degree \m\.
Then this term occurs in HS with the coefficient (d, d) ■ const which is also
non-zero for d ^ since the symmetric form ( , ) is positively definite. □

Corollaries. (1) The quantum cohomology algebra of the flag manifold
F is canonically isomorphic to

C[p , -,Pn,qi, •■•,<?n]/(A)G ,<7), -,D n (p,q)) .

(2) The total multiplicity of critical points of the functions T q in the gener-
ating family of the characteristic Lagrangian variety L is equal to ik H*(F) =
(n + 1)!

(3) rkH q = rk H*(F) if all q { ^0.

(4) The monodromy representation Z™ +1 = 7i"i(C — 0) n+1 — ► Aut(7i q )
of the Gauss-Manin connection on 7i q is unipotent and equivalent to the
action of Z n+1 on H*(F,C) generated by the multiplication operators A i— >
exp(27up fc ), k — 0, ...,n.

(5) For each A G H*(F) the function Sa has the stationary phase repre-
sentation j Tc y ex P(-^g/^) Ct; <j w tth suitable T = T(A) G H q <8> C((/l -1 )).

(6) Vice versa, the stationary phase integrals Sr admit the series expan-
sions (e pt/h s,A(T)}.

Remark. As it is shown in ||12|| , the Poincare pairing ( , ) on the quantum
cohomology algebra C\p, q]/(D (p, q), D n (p, q)) can be described by the

11

residue formula

(AB)(q)

(2%i)

1

Ajp, q)B(p, q)dp A ... A dp.,
D (p,q)...D n (p,q)

■n

Comparison of the stationary phase approximations for the integrals St with
the formal asymptotics

for solutions of the quantum cohomology differential equations shows that the
Jacobian det(dDi/dpj) computed at generic points (p,q) G L coincides, up
to a constant factor, with the Hessian det(<9 2 \F q {T) j 'dT u dT^) of the function
T q computed at the non-degenerate critical points corresponding to (p, q) . It
would be interesting to find a direct proof of this identity between the two
determinants of sizes n + 1 and n(n + l)/2 respectively.

4. The mirror conjecture. By the mirror conjecture one usually means
the profound equivalence (see e. g. f2(J, [15], discovered several years ago
on the basis of string theory, between complex and symplectic geometry in
Calabi - Yau manifolds (i.e. compact Kahler manifolds which admit non-
vanishing holomorphic volume forms). In particular, the conjecture predicts
that the quantum cohomology D-module corresponding to a Calabi - Yau
manifold X describes variations of periods of the holomorphic volume form
on another Calabi - Yau manifold Y, which has the same dimension as X
and whose Hodge diamond is mirror-symmetric to that of X.

In 1993 we suggested a generalization of the above correspondence beyond
the class of Calabi - Yau manifolds. f\ Namely, we conjectured (see |§) that
the quantum cohomology D-module of a compact symplectic 2iV-dimensional
manifold X is equivalent to the D-module generated by stationary phase
integrals J rgy exp^g/ h)u q , where (Y q , T q , oj q ) is a suitable algebraic family
of (possibly non-compact) iV-dimensional complex manifolds Y q , holomorphic
functions T q : Y q — > C and non- vanishing holomorphic iV-forms on Y q .

2 We should mention however that (earlier) E. Witten's paper jl9| can be also un-
derstood as a suggestion of a similar generalization. I am thankful to M. Atiyah who
communicated to me this point of view usually ignored by successors of [[HI .

e E 2Ji( P )/n

pe£nT* xpt

12

The following arguments spoke in favor of such a generalization.

1) In terms of the characteristic Lagrangian variety L C T*(H 2 (X) )
of the quantum cohomology D-module the Poincare pairing ( , ) on the
quantum cohomology algebra C[L] of the compact symplectic manifold X is
given by the formula

(where A is the restriction to the diagonal in L x L of the function represent-
ing Poincare-dual class of the diagonal in X x X). This formula resembles
the residue pairing

1 / / w s dyi A ... A dy N _ \ - a{y*)Kv*)
(2m) N f a[y) [y) 9^ ~ 2^ det(<9 2 ^/%%)L

in singularity theory, and in some examples, including complex projective
spaces and Grassmannians, can be indeed replaced by some residue formula.

2) Solutions sa to the differential equations hqidsA/dqi = Pi° sa arising
from the multiplication o in the quantum cohomology algebra of X admit
asymptotical approximations

{s A} B)~hW £ C p (A)e- K ^ /h [^- + 0(h)]
P eT q *nL vA(p)

resembling the stationary phase approximations

rcy„

det 1/2 (d^ q /d yi dy

jj\y*

3) A by-product of our proof JTT] for toric symplectic manifolds of Arnold's
symplectic fixed point conjecture |]J was the following multiplicative struc-
ture in the Floer homology (it is at least morally equivalent to the quantum
multiplication). A compact toric symplectic manifold X with Picard number
n can be obtained as a symplectic reduction X = C N / /T n of the linear space
by the subtorus T n C T N of the maximal torus on a generic level of the mo-
mentum map C N — > Lie*T N — > Lie*T n . Denote (wiy)^ f=i the matrix of the

13

projection Lie *T N — > Lie *T n . In the quantum cohomology algebra of X the
classes Qi, ,Qn of coordinate hyperplane divisors satisfy the multiplica-
tive relations Q™ a ...Q™ iN = q iy i = 1, ...,n, and on the other hand — can be
expressed via some basis in H 2 (X) as Qj = Pimy + ... +p n m nj,j = 1, N.
It is easy to see that the latter set of additive relations specifies the critical
set of the function T = Qi + ... + Qn restricted to the (N — n)-dimensional
complex torus Y q defined by the multiplicative relations. We arrive to the
mirror family (Y q , F q , uj q ) of the toric manifold X by introducing the holo-
morphic volume form uj q = dlogQi A ... A dlogQjy / dlogqi A ... A d\ogq n on
the torus Y q .

4) In the case of X = CP^ -1 = C N / /T 1 corresponding stationary phase
integrals

(Qi+...+Q N )/h d Q^ A ••• A d Qn

, rc{Q 1 ...Q N = q } dq
satisfy the same differential equation {hqd/dq) N S = qS as the series

( P iog 9 )/av- Q_ mod /

{ P + h) N {p + 2h) N ...{p + dhy

generating the quantum cohomology P-module of the complex projective
space (see [TO]). This confirms the mirror conjecture for CP^ -1 .

5) The above mirror family (Y q , jF q , u q ) of a toric symplectic manifold X
agrees with Batyrev's mirrors of Calabi - Yau anti-canonical hypersurfaces
X' C X. According to 0] one can construct Calabi - Yau mirror manifolds Y'
as anti-canonical divisors in the toric manifold obtained by dualization of the
momentum polyhedron for X. In fact periods of holomorphic volume forms
on Batyrev's mirrors can be expressed in terms of our mirrors (Y q , JF 9 , u q ) as
the formal Laplace transform

u q

of the stationary phase integrals. f\

3 By the way, this observation suggests a formulation of a "quantum Lefschetz hyper-
plane section theorem" relating the quantum cohomology 2?-module of a Fano manifold
X with that of the anti-canonical hypersurfaces X' C X via such a Laplace transform, i.e.
without any mentioning mirror manifolds.

14

In || we proved the mirror conjecture for Fano (Kx < 0) and Calabi -
Yau (Kx = 0) projective complete intersections. Namely, the solutions Sa
of the quantum cohomology differential equation for X given in CP N ~ l by r
equations of degrees l\ + ... + l r < N are described by the following integrals
J FcY exp^g/h) uj q : partition the variables Qi,...,Qn into r + 1 groups of
lengths lo + h + ... + l r = N and denote Go, ...,G r the total sums Q a in
each group (for example, Go = Q\ + ... + Qi , etc.). Then

Y q = {(Qi, Qn) I Qx-Qn = q, Gx(Q) = ... = G r (Q) = 1},
_ _ rflogQi A ... AdlogQ N

J- q -l* \Y q , U q - d]ogqAdGlA _ Ad Q r ■

These formulas include the traditional Calabi - Yau mirror phenomenon as
a degenerate particular case with l = and G = 0.

The very idea that the mirror phenomenon exists beyond the class of
Calabi - Yau manifolds has not attracted much attention of specialists. One
of the causes is that the results mentioned in this section do not go further
than toric complete intersections. It was the actual purpose of the project

[12] on flag manifolds started in 1993 by the author and B. Kim to improve
this situation and enlarge the supply of examples confirming the conjecture.
Although the discovered relation with Toda lattices has been analyzed in the
literature on quantum cohomology of flag manifolds (see e. g. |L7], 0, |13|),
their mirrors have escaped us so far.

In the recent paper T. Eguchi, K. Hori and C.-S. Xiong, independently
on our lecture , have arrived to a similar idea of extending the mirror con-
jecture to Fano manifolds. They illustrate the idea with the examples of
complex projective spaces, several rational surfaces, Grassmannians G^i and
G§2 and extrapolate formulas from the latter examples to general Grass-
mannians. These formulas served us as the starting point for the present
paper; suitably modified, they give rise to a construction of mirrors for the
flag manifolds. Indeed, the construction of (Y q , T q , uj q ) in Section 2 provides
a stationary phase integral representation for the solutions Sa of the dif-
ferential equations defined by the quantum cohomology algebra of the flag
manifold.

I would like to thank M. Kontsevich who brought the paper to my
attention and B. Kim for communicating his results |13| based on quantum

Toda lattices. I am also thankful to B. Kostant who explained to me that

15

generalizations of the present paper to arbitrary semi-simple Lie algebras
should intertwine mirror manifolds with Whittaker modules |[16| .

References

V.I. Arnold. Sur un propriete topologique des applications canoniques de
la mecanique classique. C. R. Acad. Sci. Paris 261 (1965), 3719 - 3722.

V.V. Batyrev. Dual polyhedra and mirror symmetry for Calabi - Yau
hyper surfaces in toric manifolds. Preprint, |alg-geom/93 10003 .

K. Behrend, Yu. Manin. Stacks of stable maps and Gromov - Witten
invariants. Preprint, 1995.

R. Dijkgraaf, E. Verlinde, H. Verlinde. Notes on topological string theory
and 2D quantum gravity. Proc. of the Trieste Spring School, 1990, M.
Green et al., eds., World Sci., Singapore, 1991.

B. Dubrovin. Geometry of 2D topological field theories. In: Springer
Lecture Notes in Math., 1620 (1996), 120 - 348.

T. Eguchi, K. Hori, C.-S. Xiong. Gravitational quantum cohomology.
Preprint, 1996.

S. Fomin, S. Gelfand, A. Postnikov. Quantum Schubert polynomials.
Preprint, 1996.

A. Givental. Equivariant Gromov - Witten invariants. Intern. Math.
Res. Notices, 1996, No. 13, 1 - 63.

A. Givental. Homological geometry and mirror symmetry. In: Proceed-
ings of the International Congress of Mathematicians, Zurich, 1994,
Birkhauser, 1995, v. 1, pp. 472 - 480.

A. Givental. Homological geometry I: projective hyper surf aces. Selecta
Math., New Series, 1 (1995), No. 2, 325 - 345.

A. Givental. A simplectic fixed point theorem for toric manifolds. In:
The Floer memorial volume. H. Hofer, C. H. Taubes, A. Weinstein, E.
Zehnder (eds.), Progress in Math. 133, Birkhauser, 1995, pp. 445 - 481.

16

A. Givental, B. Kim. Quantum cohomology of flag manifolds and Toda
lattices. Commun. Math. Phys. 168 (1995), 609 - 641.

B. Kim. Quantum cohomology of flag manifolds G/B and Toda lattices.
Preprint, |alg-geom/960700l| .

M. Kontsevich. Enumeration of rational curves via toric actions. In:
The moduli space of curves, R. Dijkgraaf, C. Faber, van der Geer (eds.),
Progress in Math. 129, Birkhauser, 1995, pp. 335 - 368.

M. Kontsevich. Homological algebra of mirror symmetry. In: Proceed-
ings of the International Congress of Mathematicians, Zurich, 1994,
Birkhauser, 1995, v. 1, pp. 120 - 139.

B. Kostant. On Whittaker vectors and representation theory. Invent.
Math. 48 (1978), 101-184.

B. Kostant. Flag manifold quantum cohomology, the Toda lattice, and
the representation with the highest weight p. Selecta Math. New Series
2 (1996), 43 - 91.

D. Morrison. Mirror symmetry and moduli spaces. In: Proceedings of
the International Congress of Mathematicians, Zurich, 1994, Birkhauser,
1995, v. 2, pp. 1304 - 1314.

E. Witten. Phases of N = 2 theories in two dimensions. Preprint,
IASSNS-HEP-93/3.

S.-T. Yau (ed.). Essays on mirror manifolds. International Press, Hong
Kong, 1992.

17